5 Mar 2026 22:37 UTC

76 points

9 comments3 min readLW link

Personality Self-Replicators

eggsyntax5 Mar 2026 20:30 UTC

173 points

51 comments10 min readLW link

Salient Directions in AI Control

Bruce W. Lee5 Mar 2026 19:38 UTC

13 points

0 comments14 min readLW link

(brucewlee.com)

Models have linear representations of what tasks they like

OscarGilg5 Mar 2026 18:44 UTC

55 points

16 comments11 min readLW link

AI Safety Has 12 Months Left

mhdempsey5 Mar 2026 16:37 UTC

40 points

10 comments6 min readLW link

(mhdempsey.substack.com)

Have Americans Become Less Violent Since 1980?

Benquo5 Mar 2026 16:11 UTC

78 points

6 comments14 min readLW link

(benjaminrosshoffman.com)

AI #158: The Department of War

Zvi5 Mar 2026 16:10 UTC

77 points

2 comments51 min readLW link

(thezvi.wordpress.com)

Investigating Self-Fulfilling Misalignment and Collusion in AI Control

Dmitrii Gusev and Vili Kohonen

5 Mar 2026 15:05 UTC

15 points

0 comments5 min readLW link

Computation, Chess, and Language in Artificial Intelligence

Bill Benzon5 Mar 2026 12:57 UTC

6 points

0 comments3 min readLW link

Vibe Coding cripples the mind

spookyuser5 Mar 2026 10:29 UTC

−12 points

4 comments4 min readLW link

Rational Chess

8495 Mar 2026 9:57 UTC

5 points

19 comments2 min readLW link

A Behavioural and Representational Evaluation of Goal-directedness in Language Model Agents

Gabriele Sarti, Raghu Arghal, ndalton, Fade Chen, Evgenii Kortukov, Calum McNamara, Angelos Nalmpantis, Moksh Nirvaan and Mario Giulianelli

5 Mar 2026 1:08 UTC

20 points

0 comments7 min readLW link

February 2026 Links

nomagicpill5 Mar 2026 1:03 UTC

6 points

2 comments7 min readLW link

(nomagicpill.substack.com)

Text Compression Can Help Secure Model Weights

Roy Rinberg4 Mar 2026 23:30 UTC

45 points

12 comments10 min readLW link

A summary of Condensation and its relation to Natural Latents

Jeremy Gillen and Daniel C

4 Mar 2026 22:22 UTC

86 points

0 comments10 min readLW link

Maybe there’s a pattern here?

dynomight4 Mar 2026 20:32 UTC

193 points

45 comments7 min readLW link

Gemini 3.1 Pro Aces Benchmarks, I Suppose

Zvi4 Mar 2026 20:10 UTC

27 points

2 comments9 min readLW link

(thezvi.wordpress.com)

Gen Z and AI: Education, Well-being, and the Labour Market

Jason Hung4 Mar 2026 19:19 UTC

9 points

0 comments11 min readLW link

Is GDP a Kind of Factory?

Benquo4 Mar 2026 18:00 UTC

59 points

3 comments10 min readLW link

(benjaminrosshoffman.com)

Make Powerful Machines Verifiable

Naci Cankaya4 Mar 2026 14:20 UTC

22 points

4 comments4 min readLW link

How a Pinky Promise once stopped a war in the Middle East.

positivesum4 Mar 2026 13:30 UTC

29 points

1 comment4 min readLW link

Split Personality Training can detect Alignment Faking

Florian_Dietz4 Mar 2026 11:49 UTC

39 points

0 comments6 min readLW link

Physics of RL: Toy scaling laws for the emergence of reward-seeking

Alex Meinke4 Mar 2026 8:12 UTC

120 points

9 comments10 min readLW link

Sacred values of future AIs

Cleo Nardo4 Mar 2026 7:47 UTC

58 points

4 comments5 min readLW link

OpenAI’s surveillance language has many potential loopholes and they can do better

Tom Smith4 Mar 2026 4:25 UTC

153 points

4 comments10 min readLW link

Mass surveillance, red lines, and a crazy weekend

Boaz Barak4 Mar 2026 4:24 UTC

35 points

26 comments5 min readLW link

Lie To Me, But At Least Don’t Bullshit

Czynski4 Mar 2026 2:20 UTC

20 points

4 comments5 min readLW link

(dangeroussincerity.substack.com)

[Question] LLM coherentization as an obvious low-hanging fruit to try?

Épiphanie Gédéon4 Mar 2026 0:59 UTC

26 points

2 comments2 min readLW link

Milder temperature makes a hell stable

Joachim Bartosik3 Mar 2026 22:25 UTC

17 points

1 comment2 min readLW link

Mass Surveillance w/ LLMs is the Default Outcome. Contracts Won’t Change That.

Logan Riggs3 Mar 2026 21:18 UTC

43 points

1 comment2 min readLW link

A Tale of Three Contracts

Zvi3 Mar 2026 20:30 UTC

45 points

3 comments16 min readLW link

(thezvi.wordpress.com)

An Alignment Journal: Coming Soon

Dan MacKinlay, JessRiedel, Edmund Lau, Daniel Murfet, Scott Aaronson, Jan_Kulveit, david reinstein, Alexander Gietelink Oldenziel and Marcus Hutter

3 Mar 2026 20:27 UTC

271 points

34 comments6 min readLW link

(blog.alignmentjournal.org)

Current activation oracles are hard to use

aryaj, Senthooran Rajamanoharan and Neel Nanda

3 Mar 2026 19:33 UTC

83 points

4 comments16 min readLW link

[Question] Question: Why is the goal of AI safety not ‘moral machines’?

Mordechai Rorvig3 Mar 2026 18:16 UTC

10 points

15 comments1 min readLW link

An Age Of Promethean Ambitions

sonicrocketman3 Mar 2026 18:09 UTC

0 points

2 comments4 min readLW link

(brianschrader.com)

White-Box Attacks on the Best Open-Weight Model: CCP Bias vs. Safety Training in Kimi K2.5

Corm3 Mar 2026 17:47 UTC

16 points

2 comments5 min readLW link

I Had Claude Read Every AI Safety Paper Since 2020, Here’s the DB

Corm3 Mar 2026 17:47 UTC

59 points

13 comments3 min readLW link

Constitutional Black-Box Monitoring for Scheming in LLM Agents

Simon Storf, richbc and Marius Hobbhahn

3 Mar 2026 17:01 UTC

28 points

0 comments2 min readLW link

LASR Labs Summer 2026 applications are open!

Brandon Riggs, Erin Robertson and LASR Labs

3 Mar 2026 15:42 UTC

25 points

0 comments3 min readLW link

AI companies and the 99% lethal autonomous weapons myth

User_Luke3 Mar 2026 12:18 UTC

−1 points

1 comment2 min readLW link

I’m confused by the change in the METR trend

Expertium3 Mar 2026 11:30 UTC

46 points

17 comments2 min readLW link

Zurich AI Safety is hiring a Director

MariusWenk, marcel.steimke, Lukas Fluri and alex.lloyd

3 Mar 2026 10:29 UTC

21 points

0 comments3 min readLW link

Game Recognizes Game

eva_3 Mar 2026 10:09 UTC

105 points

15 comments12 min readLW link

Monday AI Radar #15

Against Moloch3 Mar 2026 5:23 UTC

13 points

1 comment7 min readLW link

Memory Decoding Journal Club: Engram cell connectivity as a mechanism for information encoding and memory function

Devin Ward3 Mar 2026 1:32 UTC

3 points

0 comments1 min readLW link

In-context learning of representations can be explained by induction circuits

Andy Arditi2 Mar 2026 23:58 UTC

51 points

0 comments9 min readLW link

(iclr-blogposts.github.io)

Single Direction vs Low-Rank Refusal in Small LLMs

IvanC2 Mar 2026 23:14 UTC

12 points

0 comments8 min readLW link

Being ambitious in soulful altruism

pandamonium2 Mar 2026 21:15 UTC

5 points

0 comments4 min readLW link

Notes on the “Heart of Darkness”

dominicq2 Mar 2026 20:11 UTC

5 points

0 comments4 min readLW link

(sundaystopwatch.eu)

[Question] Can LLM chat be less prolix?

jbash2 Mar 2026 19:54 UTC

21 points

9 comments2 min readLW link